A New Infrastructure Demand Model for Urban Business and Leisure Hubs - a case study of Taichung

206 

Full text

(1)

Hsin-tzu Ho Queens’ College

A dissertation submitted to the University of Cambridge for the Degree of Doctor of Philosophy

Department of Architecture

September 2016

This dissertation is submitted for the degree of Doctor of Philosophy

A New Infrastructure Demand Model for Urban

Business and Leisure Hubs - a case study of Taichung

(2)
(3)

i Abstract

A New Infrastructure Demand Model for Urban Business and Leisure Hubs - a case study of Taichung

By Hsin-tzu Ho

Over the last few decades there has been a gradual transformation in both the spatial and temporal patterns of urban activities. The percentage share of non-discretionary travel such as morning rush-hour commuting has been declining with the increased income level. Discretionary activities appear to rise prominently in urban business and leisure hubs, attracting large volumes of crowds which in turn imply new and changed demand for building floorspace and urban infrastructure.

Despite impressive advances in the theories and models of infrastructure demand forecasting, there appear to be an apparent research gap in addressing the practical needs of infrastructure planning in and around those growing urban activity hubs. First, land use and transport interaction models which have to date been the mainstay of practical policy analytics tend to focus on non-discretionary activities such as rush-hour commuting. Secondly, the emerging activity based models, while providing significant new insights into personal, familial activities, especially the discretionary travel, are so data hungry and computing intensive that they have not yet found their roles in practical policy applications.

This dissertation builds on the insights from above schools of modelling to develop a new approach that addresses the infrastructure planning needs of the growing urban hubs while keeping the data and computing realistic in medium to high income cities. The new model is designed based on an overarching hypothesis that considerable efficiency and welfare gains can be achieved in the planning and development of urban business and leisure hubs if the infrastructure provisions for discretionary and non-discretionary activities can be coordinated. This is a research theme that has been little explored in current literature.

The new infrastructure demand forecasting model has been designed with regard to the above hypothesis and realistic data availability, including those emerging online.

(4)

ii

The model extends the framework of land use transport interaction models and aim to provide a practical modelling tool. Land use changes are accounted for when testing new infrastructure investment initiatives and especially the road and public transport loads are assessed throughout all time periods of a working day.

The new contribution to the modelling methodology includes the extension to the land use transport interaction framework, the use of social media data for estimating night market activity distribution and a rapid estimation of road traffic speeds from Google directions API, and model validation. Another new contribution is the understanding of the nature and magnitude of future infrastructure demand through assessing three alternative land use scenarios: (1) business as usual, (2) inner city regeneration for a major business hub around the night market, and (3) dispersed suburban growth with distant subcentres. The model is able to assess the implications for future infrastructure demand and user welfare through discerning the distinct discretionary and non-discretionary activity patterns.

Key words: Discretionary travel demand, 24-hour traffic modelling, social media data, congested link speed estimation, integrated land use and transport model

(5)

iii Declaration

This dissertation is the result of my own work and includes nothing which is the outcome of work done in collaboration except as declared in the Preface and specified in the text.

It is not substantially the same as any that I have submitted, or, is being concurrently submitted for a degree or diploma or other qualification at the University of Cambridge or any other University or similar institution except as declared in the Preface and specified in the text. I further state that no substantial part of my dissertation has already been submitted, or, is being concurrently submitted for any such degree, diploma or other qualification at the University of Cambridge or any other University of similar institution except as declared in the Preface and specified in the text

(6)

iv Acknowledgement

This dissertation would not have been possible without the guidance of Dr Ying Jin, my supervisor. His invaluable guidance, donation of time spent tutoring me throughout the year and patience has made it possible for me to complete this dissertation.

I would like to express my sincere gratitude to all those who help me during the time of writing this dissertation. I wish to extend my thanks to the Steve Denman who provides me with his GIS and Python expertise which are indispensable for this research. I also want to thank the WSP group for providing the MEPLAN software package for my research. Also, THI consultant, Professor Lee from Feng Chia University and Chung-Den Huang from Taichung city government all provided me with Taichung data with open attitude which enabled a smooth operation of data collection process.

Finally my deepest thanks go to my dearest parents, sister and my partner, without their encouraging and support, this dissertation cannot be completed. Very special thanks should go to all my dear friends in the UK, with their love and company, I could do this work.

(7)

1

Contents

Chapter 1 INTRODUCTION ... 10

1.1 Objectives ... 10

1.2 Methodology ... 11

1.3 Choice of case study ... 12

1.4 Summary ... 12

Chapter 2 LITERATURE REVIEW ... 14

2.1 Overview... 14

2.2 Conventional four-stage transport models ... 14

2.2.1 Aggregate Models ... 15

2.2.2. Disaggregate trip-based models ... 17

2.3 Activity-based models ... 19

2.4 Land use and transport interaction models ... 20

2.5 A case study of land use and transport model software MEPLAN ... 24

2.5.1 Overall structure and operation of the model ... 25

2.5.2 The land use module (LUS) ... 26

2.5.3 The land use/transport interface module (FRED) ... 27

2.5.4 The Transport module (TAS) ... 28

2.6 Case study 2: SIMULACRA ... 28

2.7 Critical summary ... 31

Chapter 3 METHODOLOGY ... 33

3.1 Methodological framework ... 33

3.2 Land use ... 35

3.2.1 Conventional activities ... 36

3.2.2 Night market activities ... 37

3.3 Conversion of production-attraction matrices to OD matrices ... 46

3.4 Transport Models ... 47

3.4.1 Modal choice models ... 48

3.4.2 Networks and assignment models ... 48

3.4.3 Intrazonal links ... 50

3.5 Scenario tests ... 52

Chapter 4 DATA, MODEL CALIBRATION AND VALIDATION ... 54

(8)

2

4.2 Data sources ... 56

4.2.1 Feng-chia survey ... 58

4.2.2 Social media data ... 62

4.2.3 Web data extraction ... 66

4.3 Land use model ... 66

4.3.1 Conventional activities ... 66

4.3.2 Night market activities ... 68

4.4 Transport model specifications ... 78

4.4.1 Passenger travel demand segments ... 79

4.4.2 Transport network... 81

4.4.3 Definition of transport modes ... 86

4.5 Model Calibration ... 87

4.5.1 Calibrating spatial distribution and modal split for conventional activities ... 87

4.5.2 Calibrating spatial distribution and modal split for night market activities ... 90

4.6 Model Implementation ... 92

Chapter 5 SCENARIO TESTS ... 94

5.1 Exogenous Demographic Assumptions ... 95

5.2 The BAU Scenario ... 97

5.2.1 Travel demand ... 97

5.2.2 Link traffic volumes under BAU 2041 ... 104

5.2.3 Link traffic volumes comparison between BAU 2041 and Existing 2013 ... 108

5.3 The RaSnAS Scenario ... 112

5.3.1 Travel demand ... 113

5.3.2 Link traffic volumes under RaSnAS 2041 ... 122

5.3.3 Link traffic volumes comparison between RaSnAS 2041 and BAU 2041 ... 126

5.4 The RS Scenario ... 132

5.4.1 Travel demand between the zones ... 133

5.4.2 Link traffic volumes under RS 2041 ... 138

5.4.3 Link traffic volumes comparison between RS 2041 and BAU 2041 ... 141

5.5 Sensitivity analysis... 147

Chapter 6 CONCLUSIONS ... 157

6.1 Findings and insights ... 157

6.2 Strengths, weaknesses and further development of the model ... 160

(9)

3

Appendix 1 MODEL ZONING SYSTEM ... 168

Appendix 2 NETWORK LINK TYPES ... 174

Appendix 3 CONVERSION RATE ... 180

Appendix 4 TRANSPORT MODES DEFINITION ... 182

Appendix 5 USER MODE CONSTANTS ... 183

Appendix 6 LAND USE ASSUMPTION ... 187

Appendix 7 TRAVEL DEMAND OF FENG-CHIA PRECINCT AND THE REST OF THE STUDY AREA ... 190

(10)

4 List of Figure

Figure 2-1 Complexity of functional linkages in urban system dynamics ... 20

Figure 2-2 Generalized structure of an operational LUTI model ... 21

Figure 2-3 Chronological development of LUTI models ... 23

Figure 2-4 Dynamic operation of the model ... 25

Figure 2-5 Typical structure and operation of a MEPLAN model ... 26

Figure 2-6 The SIMULACRA model structure. ... 29

Figure 2-7 SIMULACRA framework overall sequence of sub-models ... 30

Figure 3-1 Proposed model structure ... 34

Figure 4-1 Vehicular flow recorded near Feng-Chia night market in Taichung ... 54

Figure 4-2 Traffic at Feng-chia night market at 8:30(Left) and 22:00(Right) ... 54

Figure 4-3 Zoning system for the case study area in Taichung ... 56

Figure 4-4 Model structure and data for the Calibration Year 2013 ... 58

Figure 4-5 City of residence based on Feng-chia survey (2014) ... 59

Figure 4-6 Share of residents and tourists in some commercial centres in Taichung based on Chen and Wang (2009) ... 60

Figure 4-7 Share of residents and tourists in some commercial centres in Taichung based on scraped IPeen data (2015) ... 60

Figure 4-8 Modal share comparison ... 61

Figure 4-9 Age share comparison ... 62

Figure 4-10 Number of IPeen users by level ... 63

Figure 4-11 Number of comments by user level ... 63

Figure 4-12 Average number of comments by user level ... 64

Figure 4-13 Average spend per shop ... 64

Figure 4-14 Tourists’ residence based on food reviews in TripAdvisor ... 65

Figure 4-15 Shares of age group as constraints for calibration of spatial allocation model of night market activity ... 70

Figure 4-16 Routes for estimating congested speeds on the urban roads with median interference ... 82

Figure 4-17 Public transport provisions in Taichung ... 85

Figure 4-18 Modal share for HBW ... 89

Figure 4-19 Modal share for HBE ... 89

Figure 4-20 Modal share for HBO ... 90

Figure 4-21 Modal share for NHB ... 90

Figure 4-22 Calibration results of tourists’ modal split ... 92

Figure 5-1 Projected change in households 2013- 2041 in Taichung ... 96

Figure 5-2 Projected change in population 2013- 2041 in Taichung ... 96

Figure 5-3 Link load/capacity ratio in AM peak hour for BAU 2041 ... 105

Figure 5-4 Link load/capacity ratio in Interpeak hour for BAU 2041 ... 105

Figure 5-5 Link load/capacity ratio in PM peak hour for BAU 2041 ... 106

Figure 5-6 Link load/capacity ratio in Evening peak hour for BAU 2041 ... 106

Figure 5-7 Maximum link load/capacity ratio > 1 by peak hour for BAU 2041 ... 107

Figure 5-8 Maximum link load/capacity ratio between 0.5 and 1 by peak hour for BAU 2041 ... 107

(11)

5

Figure 5-9 Maximum hourly flow ratio between BAU 2041 and Existing 2013 scenario

... 108

Figure 5-10 AM period flow when the ratio is larger than 1 ... 109

Figure 5-11 Interpeak period flow when the ratio is larger than 1 ... 109

Figure 5-12 PM period flow when the ratio is larger than 1 ... 110

Figure 5-13 Evening period flow when the ratio is larger than 1 ... 110

Figure 5-14 Link entropy between Existing 2013 and BAU 2041 scenario ... 111

Figure 5-15 Location of Shui-nan airport regeneration site and Feng-chia precinct in Taichung ... 112

Figure 5-16 Comparison of passenger trip volume to zone 60 ... 121

Figure 5-17 Passenger trip volume to zone 60 in airport site under RaSnAS 2041 ... 121

Figure 5-18 Link load/capacity ratio in AM peak hour for RaSnAS 2041 ... 123

Figure 5-19 Link load/capacity ratio in Interpeak hour for RaSnAS 2041 ... 123

Figure 5-20 Link load/capacity ratio in PM peak hour for RaSnAS 2041 ... 124

Figure 5-21 Link load/capacity ratio in Evening peak hour for RaSnAS 2041 ... 124

Figure 5-22 Maximum link load/capacity ratio > 1 by peak hour for RaSnAS 2041 ... 125

Figure 5-23 Maximum link load/capacity ratio between 0.5 and 1 by peak hour for RaSnAS 2041... 125

Figure 5-24 Link flow change between RaSnAS 2041 and BAU 2041 in AM peak hour 126 Figure 5-25 Link flow change between RaSnAS 2041 and BAU 2041 in Interpeak hour ... 127

Figure 5-26 Link flow change between RaSnAS 2041 and BAU 2041 in PM peak hour 127 Figure 5-27 Link flow change between RaSnAS 2041 and BAU 2041 in Evening peak hour ... 128

Figure 5-28 Maximum link flow change between RaSnAS 2041 and BAU 2041 ... 128

Figure 5-29 Ratio of maximum hourly link flow between RaSnAs 2041 and BAU 2041 scenario ... 129

Figure 5-30 AM period flow when maximum hourly link flow RaSnAS 2041 > BAU 2041 ... 129

Figure 5-31 Interpeak period flow when maximum hourly link flow RaSnAS 2041 > BAU 2041 ... 130

Figure 5-32 PM period flow when maximum hourly link flow RaSnAS 2041 > BAU 2041 ... 130

Figure 5-33 Evening period flow when maximum hourly link flow RaSnAS 2041 > BAU 2041 ... 131

Figure 5-34 Link entropy between BAU 2041 and RaSnAS 2041 scenario ... 132

Figure 5-35 Location of subcentres in Taichung ... 133

Figure 5-36 Comparison of passenger trip volume to destination zone 49 ... 137

Figure 5-37 Passenger trip volume to destination zone 49 under RS 2041 ... 137

Figure 5-38 Link locad/capacity ratio in AM peak hour for RS 2041 ... 138

Figure 5-39 Link load/capacity ratio in Interpeak hour for RS 2041 ... 139

Figure 5-40 Link load/capacity ratio in PM peak hour for RS 2041 ... 139

Figure 5-41 Link load/capacity ratio in Evening peak hour for RS 2041 ... 140

(12)

6

Figure 5-43 Maximum link load/capacity ratio between 0.5 and 1 by peak hour for RS

2041 ... 141

Figure 5-44 Link flow change between RS 2041 and BAU 2041 in AM peak hour ... 142

Figure 5-45 Link flow change between RS 2041 and BAU 2041 in Interpeak hour ... 142

Figure 5-46 Link flow change between RS 2041 and BAU 2041 in PM peak hour ... 143

Figure 5-47 Link flow change between RS 2041 and BAU 2041 in Evening peak hour 143 Figure 5-48 Maximum link flow change between RS 2041 and BAU 2041 ... 144

Figure 5-49 Ratio of maximum hourly link flow between RS 2041 and BAU 2041 ... 144

Figure 5-50 AM period flow when maximum hourly link flow RS 2041 > BAU 2041.... 145

Figure 5-51 Inter peak period flow when maximum hourly link flow RS 2041 > BAU 2041 ... 145

Figure 5-52 PM period flow when maximum hourly link flow RS 2041 > BAU 2041 .... 146

Figure 5-53 Evening period flow when maximum hourly link flow RS 2041 > BAU 2041 ... 146

Figure 5-54 Link entropy between BAU 2041 and RS 2041 scenario ... 147

Figure 5-55 Link entropy between Existing 2013 and BAU 2041 scenario, with concentration parameters increased by 25% ... 149

Figure 5-56 Link entropy between BAU 2041 and RaSnAS 2041 scenario, with concentration parameters increased by 25% ... 149

Figure 5-57 Link entropy between BAU 2041 and RS 2041 scenario, with concentration parameters increased by 25% ... 150

Figure 5-58 Link entropy between Existing 2013 and BAU 2041 scenario, with concentration parameters decreased by 25% ... 150

Figure 5-59 Link entropy between BAU 2041 and RaSnAS 2041 scenario, with concentration parameters decreased by 25% ... 151

Figure 5-60 Link entropy between BAU 2041 and RS 2041 scenario, with concentration parameters decreased by 25% ... 151

Figure 5-61 Link entropy between Existing 2013 and BAU 2041 scenario, with demand coefficient decreased to 0.5 ... 153

Figure 5-62 Link entropy between BAU 2041 and RaSnAS 2041 scenario, with demand coefficient decreased to 0.5 ... 153

Figure 5-63 Link entropy between BAU 2041 and RS 2041 scenario, with demand coefficient decreased to 0.5 ... 154

Figure 5-64 Link entropy between Existing 2013 and BAU 2041 scenario, with weighting parameters changed to 0.7 and 0.3 ... 155

Figure 5-65 Link entropy between BAU 2041 and RaSnAS 2041 scenario, with weighting parameters changed to 0.7 and 0.3 ... 156

Figure 5-66 Link entropy between BAU 2041 and RS 2041 scenario, with weighting parameters changed to 0.7 and 0.3 ... 156

(13)

7 List of Table

Table 2-1 Variables used in SIMULACRA ... 30

Table 4-1 Demand coefficient of various conventional activities by household type ... 67

Table 4-2 Parameters for estimating the attractiveness of a zone ... 67

Table 4-3 Estimated demand coefficient for night market activity for residents ... 69

Table 4-4 Demand coefficients and concentration parameters for home based demand71 Table 4-5 Demand coefficients and concentration parameters for non-home based demand ... 71

Table 4-6 Estimated number of guests per day for all establishments by type ... 73

Table 4-7 Examples of data from Tourism Bureau of Taichung and TaiwanStay ... 73

Table 4-8 Number of guests per night by type of accommodation establishment ... 74

Table 4-9 Examples of scraped data from TripAdvisor, IPeen and Foursquare and the corresponding model zone ... 75

Table 4-10 Some zones with the number of accommodation establishments listed on TripAdvisor, IPeen and Foursquare ... 75

Table 4-11 Data from Tourism Bureau of Taichung and the derived number of guests . 76 Table 4-12 Some zones with the number of tourist by accommodation establishments type ... 76

Table 4-13 Examples of Airbnb data and the matching model zone ... 77

Table 4-14 Some zones with the estimated number of Airbnb guests per night ... 77

Table 4-15 Example of estimated number of tourists staying at accommodation establishments ... 78

Table 4-16 Number of tourist staying with friends and family by zone... 78

Table 4-17 Flow definition by type of activity by time period ... 80

Table 4-18 Proportions of flows to activities by departure time period ... 80

Table 4-19 Proportions of flows from activities by departure time period ... 80

Table 4-20 Comparison of travel time and distance between model results and Google Direction data ... 83

Table 4-21 Definition of intrazonal distance bands and link types ... 84

Table 4-22 Link speeds by intrazonal band (km/h) ... 84

Table 4-23 Value of time by travel demand segment ... 86

Table 4-24 Concentration parameters by type ... 88

Table 4-25 Distance band distribution by trip purpose ... 88

Table 4-26 Calibration results of residents’ spatial distribution and modal split for night market activities ... 91

Table 4-27 Calibration results of tourists’ spatial distribution and modal split ... 91

Table 5-1 Travel demand by trip purpose by mode under Existing 2013 scenario ... 98

Table 5-2 Travel demand by time period by mode under Existing 2013 scenario ... 99

Table 5-3 Travel demand by trip purpose by mode under BAU 2041 scenario ... 100

Table 5-4 Travel demand comparison by trip purpose by mode between BAU 2041 and Existing 2013 ... 101

Table 5-5 Travel demand by time period by mode under BAU 2041 scenario ... 102

Table 5-6 Travel demand comparison by time period by mode between BAU 2041 and Existing 2013 ... 103

(14)

8

Table 5-7 Land use change in the airport ... 112 Table 5-8 Travel demand comparison by trip purpose by mode between RaSnAS 2041 and BAU 2041 ... 114 Table 5-9 Travel demand comparison by time period by mode between RaSnAS 2041 and BAU 2041 ... 115 Table 5-10 Comparison of travel demand by activity by mode: BAU 2041 vs RaSnAS 2041 ... 118 Table 5-11 Travel demand relating to Feng-chia precinct under Existing 2013 scenario ... 119 Table 5-12 Travel demand relating to Feng-chia precinct under BAU 2041 scenario in relation to Existing 2013 ... 119 Table 5-13 Travel demand relating to Feng-chia precinct under RaSnAS 2041 scenario in relation to BAU 2041 ... 119 Table 5-14 Land use change in the subcentres compared to BAU 2041 ... 133 Table 5-15 Travel demand comparison by trip purpose by mode between RS 2041 and BAU 2041 ... 134 Table 5-16 Travel demand comparison by time period by mode between RS 2041 and BAU 2041 ... 135 Table 5-17 Comparison of travel demand by activity by mode: BAU 2041 vs RS 2041 136 Table 5-18 Weighted link entropy comparison with changes in concentration

parameters... 148 Table 5-19 Weighted link entropy comparison with change in demand coefficient ... 152 Table 5-20 Weighted link entropy comparison with change in weighting parameteres ... 155

(15)

9 List of Abbreviations

CECI China Engineering Consultants Incorporated DORTS Department of Rapid Transit Systems

GIS Geographic information system GLA Greater London Authority HBE Home Based Education HBO Home Based Other HBW Home Based Work

LUTI Land use and transport interaction model MOTC Ministry of Transportation and Communications NHB Non Home Based

(16)

10

Chapter 1 INTRODUCTION

Over the last few decades there has been a gradual transformation in both the spatial and temporal patterns of urban activities. While the total amount of movements of people and goods rises with income, the percentage share of non-discretionary travel such as morning rush-hour commuting has been declining. Urban business and leisure hubs in the cities that cater for a wide variety of discretionary activities such as impromptu business, social and leisure gatherings appear to rise prominently, attracting large volumes of crowds which in turn imply new and changed demand for building floorspace and urban infrastructure.

In most cities in the world, while the total amount of movements of people and goods rising with income, the percentage share of non-discretionary travel has been in decline. In London, for instance, work related travel (both commuting and on work business) now accounts for only 28.8% of trips made by Londoners on an average weekday while leisure trips have increased by 44% over the period between 2005/2006 and 2013/2014 and many of this increased travel demands come from people who undertake night time activities (Transport for London, 2014; 2015). In many cities, business and leisure hubs have brought prosperity and vitality to the urban life and contributed to improved productivity and quality of life due to the urban agglomeration effects. However, those urban hubs are usually located in the densest parts of the cities so any rise in activities and travel in those areas may result in congestion and overcrowding which may translate into high property and goods prices and even inefficiency and social exclusion in the long term if these issues are not addressed properly in the time of need. The need for infrastructure expansion and enhancement for addressing these issues is obvious. However, they are costly and difficult to achieve in those dense urban areas without careful coordination and suitable planning tool.

1.1 Objectives

Given the increasing evening congestion problems shown in big urban areas and due to current lack of modelling capacity for coordinating discretionary and non-discretionary activities, this dissertation aims to investigate the cost-effective development of practical planning to this end. Thus, the specific objectives of the research are to:

(1) carry out a comprehensive review of the existing methods for forecasting travel demand,

(17)

11

(2) construct a model aimed at improving current approach to forecasting evening travel, (3) to implement the model for a case study area with active night economy, and

(4) identify areas of methodological and data improvements to make adequate quantification where such evidence is lacking

1.2 Methodology

This new tool is built on the MEPLAN framework, as a new integrated land use and transport model whose model structure has been extended to accommodate additional modelling dimensions that have rarely featured in existing models. More specifically, the model incorporates new components for both discretionary and non-discretionary travel demand over all time periods of the day. To overcome the data problems that have hampered the previous attempts of a similar kind, the model makes use of new social media data which serves as an input for the land use module of the integrated model. This makes it possible to consider the interactions between transport and land use activities that underlie the rarely modelled discretionary activities. In addition to the social media data, our model makes a full use of the Population Census data (Taiwan Directorate-General of Budget, Accounting and Statistics, 2010), government statistics for tourism industry (Tourism Bureau of Taichung, 2014), Taichung metropolitan area road network planning report (China Engineering Consultants Incorporated [CECI], 2009) and the transport network and trip ends data from the TransCAD model accompanying the report, Taichung household travel survey (Department of Rapid Transit Systems [DORTS], 2016), Google Map Directions, and Feng-chia survey (Lee, Lin, & Hsieh, 2015) in the case study area.

It should be noted that although this dissertation only makes use of social media data to estimate discretionary travel demand at the evening and night times due to data limitations and time constraints, the data collection and analysis techniques and the model estimation methods for discretionary travel are transferrable to estimate discretional travel occurring during other time periods, such as inter peak hours. Also, due to the time and data constraints, the travel demand is estimated based on projected land use activity distributions rather than a comprehensive land use model. This is to say that the case study focuses on examining the traffic and transport issues for this dissertation without exploring the feedback of transport improvement on land use patterns. Notwithstanding the choice of model specification for the case study, the wider

(18)

12

economic and land use factors are included in the model design in such a way that the assessment of wider economic and social impacts can be incorporated as part of further work.

1.3 Choice of case study

Night time activities continue to grow in many parts of the world. Many Asian cities, in particular, have the tradition that the residents go to the night market as part of social life. Some of the night markets even transform into major tourist spots of an international reputation, often supposing the fame of major centres of night life in Europe or North America. For our study, choosing a case study which is informative regarding forecasting discretional travel during the non-peak modelling period is important.

Taichung is the third largest city in Taiwan among Taiwan's five special municipalities with a population of 2,720,000 people and it is in many ways a typical example of a large number of medium to high income provincial cities in the world. It boasts an arguably the biggest night market destination in Taichung and is recorded to be the most visited tourist attraction by international tourists in Taiwan (Taiwan Tourism Bureau, 2015a). Other night time activities dotted around the city attracts both local residents and tourists across East and South East Asia. By contrast, in the new urban plan to 2041, which anticipates a significant population and income growth, there has been little effort to plan or coordinate the provision of the urban infrastructure of the new city centre, subcentres and the night market areas. To this end, Taichung is chosen to be the case study area of the research.

1.4 Summary

This thesis is constructed as following:

1. Chapter 1 gives the general background and introduction; 2. Chapter 2 offers a literature review;

3. Chapter 3 describes the methodology of land use and transport model, model calibration process;

4. Chapter 4 provides a case study of operationalising the land use and strategic transport model for the case study area of Taichung, including the social media data collection and application, congested link speed estimation using Google MAP Directions API, and the examination of the networks;

(19)

13

5. Chapter 5 explores the application of the model and present the results of the Existing 2013, BAU 2041, RaSnAS 2041 and RS 2041 scenarios;

(20)

14

Chapter 2 LITERATURE REVIEW

2.1 Overview

Most metropolitan planning organisations employ the conventional trip based four step model (de Dios Ortuzar & Willumsen, 2011). In this type of model, trips outside the AM peak period are usually not considered and therefore leisure trips tend to be largely omitted from the model. Another approach, the activity-based approach, which has been the result of increasing interest in travel behaviour research, considers travel needs as a result of an individual’s or a firm’s desire to participate in activities at different geographical locations at different times (Jones et al., 1990; Bhat, 2012). This kind of approach simulates activity-chains rather than trips and thus leisure trips made in a day are incorporated explicitly into the activity-chains.

The above two approaches are not able to capture the travel behavioural change in the long term because the models do not consider the impact of the transport system on land use which may affect people’s decisions on housing and a firm’s decision to locate the business and eventually drive a shift in people’s choice in travel (Echenique, 2004). Meanwhile the increasing use of social media and the use of smart phones, offer us a new way to observe the dynamics of urban transport and land use. Social media has opened up opportunities to collect crowd-sourced information on individuals’ footprints relating to leisure activities that are not easily captured in traditional household travel surveys. Traffic data collected by GPS-enabled smart phones, prevalent in most cities nowadays, provides speed information on large numbers of roads which is again difficult to obtain with restricted resources.

The following sections provide a review of urban models to form a basis for modelling enhancement in the context of this research. Firstly, travel demand models will be reviewed and followed by the most relevant operational land use and transport interaction models.

2.2 Conventional four-stage transport models

The fundamentals of transport modelling were developed during the 1950s. That period had seen rapid increases in car use followed by major investments in new road infrastructure. This resulted in the development of an aggregate approach called the sequential four-step transport model for evaluating performance of transportation

(21)

15

systems and large-scale transport infrastructure projects. This approach starts by dividing a study area into zones. The size and number of the zones is determined based on the modelling purpose and the precision required. For each zone, base year data on demography such as the size of different population types and land use (economic activity) such as employment, shopping floor space, educational institutions and recreational facilities is needed. Also, a network which contains information on transport supply is needed. The first step is trip generation which determines the total number of trips generated from and attracted to each zone based on the demographics and land use mentioned earlier. The next step is the distribution of trips between zones which gives a trip matrix. The following step is a modal split which involves allocating trips in the matrix to different transport modes. Lastly, the trip assignment stage calculates how the trips by each transport mode will distribute onto their corresponding network.

During the 1970’s transport planning shifted focus from global infrastructure developments to the travel needs of individuals. Disaggregate models such as disaggregate trip based demand models and activity based travel demand models were developed in response to the shift. Disaggregate models flourished during the 1980’s and 1990’s and have been applied in many projects around the world in the past 20 years. Nowadays, more disaggregate versions of four-step model continuously dominate the process of travel demand modelling. The following sections describe these models in detail.

2.2.1 Aggregate Models 1. Gravity model

Aggregate models are the earliest travel demand models employing simple mathematical models, such as a gravity model. The number of trips generated from a zone was considered to be proportional to the total trip ends in the origin zone and the number of trips attracted to a zone was considered to be proportional to the total trip ends in the destination zone. Moreover, the number of trips between zones was determined by the inter-zonal impedance, i.e. the disincentive to travel as distance (time) or cost increases. Generally, the gravity model can be expressed as:

(22)

16 Tij = α OiDj f (cij)

Where

Tij is the number of trips between origin zone i and destination zone j Oiis the total trip ends at origins

Dj is the total trip ends at destinations α is a proportionality factor

f ( cij) is a generalised function of the travel cost with one or more parameters for calibration.

2. Entropy maximising technique

The derivation of the gravity model from principles of entropy maximisation (Wilson, 1970) was a major accomplishment and formed the basis for many of the allocation mechanisms within spatial interaction.

Wilson proposed an approach which considers the location of services activities expressed by: 𝐹𝑖𝑗 = 𝑃𝑖𝑊𝑗 𝛼𝑒−𝛽𝑐𝑖𝑗 ∑ 𝑊𝑙𝛼𝑒−𝛽𝑐𝑖𝑙 𝑙 Where

𝐹𝑖𝑗 represents the flows (monetary flows or number of consumer) between zone i (origin, residential zone) and zone j (destination, in which the service is located)

𝑃𝑖 is the number of consumers living in the zone i,

𝑊𝑗𝛼 represents the commercial attractiveness of the zone j, 𝑐𝑖𝑗 is the transport cost from zone i to zone j,

α and β are modifying parameters

When flows throughout the agglomeration are defined, we obtain the demand 𝐷𝑗, induced by all the agglomeration in zone j:

𝐷𝑗 = ∑ 𝐹𝑖𝑗

(23)

17

In others models, attractiveness of a zone j is an exogenous variable, measured by number of services offered in zone j, commercial surfaces and etc. This factor defines supply of services, and the model defines the distribution of the flows of consumers in function of this supply (the distribution of the supply is given "a priori"). The great interest of the Wilson's model is to consider the supply as an endogenous variable and to model its evolution through time. It assumes that the suppliers of services are interested by the distribution of the demand, and try to adapt their behaviour to it. The system tend to come to an equilibrium between production costs, function of attractiveness 𝑊𝑗 and income products, function of demand 𝐷𝑗.

2.2.2. Disaggregate trip-based models

This group of models explains the connection that exists between the characteristics of locations and the behaviour of decision maker. An individual is associated with a utility function (𝑼𝒏𝒊), regarding the attributes of location and the properties displayed by the decision maker. Each decision maker has to make his decision from a discrete set of alternative choice options and will settle on the option with the highest utility. Since not all attributes of the location and the decision maker are observable, a random measuring strict utility (𝑽𝒏𝒊), the fixed and measurable attributes of utility; and the other dealing

with stochastic utility ( 𝜺𝒏𝒊), an error or disturbance term that reflects the unobserved attributes of a given decision. And the total utility of any alternative i is expressed by the sum of observed and unobserved components. The utility function can be expressed as follows:

𝑈𝑛𝑖= 𝑉𝑛𝑖+ 𝜀𝑛𝑖 = 𝛽𝑍𝑛𝑖+ 𝜀𝑛𝑖

Where

Uni is the utility for the person n to choose alternative i

Vni is the fixed and measurable attributes of utility for the person n to choose i

εni captures the impact of all unobserved factors that affect the person’s choice.

Zni = Z (Xni, Sn) is a vector of observed variables where Xni are attributes of the alternative

i and Sn are attributes of the person n (Xni and Sn may interact with each other)

β is a corresponding vector of coefficients of the observed variables

Based on the above theory, the discrete choice model is derived and the choice probability can be express by the following equation:

(24)

18 𝑃𝑛𝑖 = 𝑃𝑟𝑜𝑏(𝑈𝑛𝑖 > 𝑈𝑛𝑗)

= 𝑃𝑟𝑜𝑏 (𝛽𝑍𝑛𝑖+ 𝜀𝑛𝑖 > 𝛽𝑍𝑛𝑗+ 𝜀𝑛𝑗)

= 𝑃𝑟𝑜𝑏 ( 𝜀𝑛𝑗− 𝜀𝑛𝑖 < 𝛽𝑍𝑛𝑖− 𝛽𝑍𝑛𝑗), ∀ 𝑗 ≠ 𝑖

Where

𝑃𝑛𝑖 is the probability for the person n to choose alternative i 𝑈𝑛𝑖 is the utility for the person n to choose alternative i 𝑈𝑛𝑗 is the utility for the person n to choose alternative j

β is a corresponding vector of coefficients of the observed variables

𝑍𝑛𝑖 = Z ( 𝑋𝑛𝑖, 𝑆𝑛) is a vector of observed variables where Xni are attributes of the alternative i and Sn are attributes of the person n (Xni and Sn may interact with each other)

𝑍𝑛𝑗 = Z ( 𝑋𝑛𝑗, 𝑆𝑛) is a vector of observed variables where Xnj are attributes of the alternative j and Sn are attributes of the person n (Xnj and Sn may interact with each other)

εni captures the impact of all unobserved factors that affect the person n’s choice of alternative i

εnj captures the impact of all unobserved factors that affect the person n’s choice of

alternative j

Different choice models arise from different distributions of 𝜺𝒏𝒊 (for all i). The most popular form of discrete choice model is logit model which means the unobserved utility is distributed extreme value. The logit choice probability can be expressed as:

𝑃𝑛𝑖 = 𝑒𝑉𝑛𝑖

∑ 𝑒j 𝑉𝑛𝑗= 𝑒𝛽𝑍𝑛𝑖 ∑ 𝑒j 𝛽𝑍𝑛𝑗

Where, as defined above

𝑃𝑛𝑖 is the probability for the person n to choose alternative i

Vni is the fixed and measurable attributes of utility for the person n to choose i Vnj is the fixed and measurable attributes of utility for the person n to choose j

(25)

19

𝑍𝑛𝑖 = Z ( 𝑋𝑛𝑖, 𝑆𝑛) is a vector of observed variables where Xni are attributes of the alternative i and Sn are attributes of the person n (Xni and Sn may interact with each other)

𝑍𝑛𝑗 = Z ( 𝑋𝑛𝑗, 𝑆𝑛) is a vector of observed variables where Xnj are attributes of the alternative j and Sn are attributes of the person n (Xnj and Sn may interact with each other)

β is a corresponding vector of coefficients of the observed variables

The major limitation of these models was the missing linkages between trips. As a result, different modes of transport could potentially be assigned to a home-to-work trip and its return trip from work.

2.3 Activity-based models

Activity—based model offered an alternative approach by representing the demand for travel as derived from the need to conduct activities separated in space and time (Axhausen & Gärling, 1992). This type of model place primary emphasis on activity participation and patterns of activities made subject to the interdependencies with other household members within the constraint of space and time (Hägerstrand, 1970 and Chapin, 1974). Thus, this type of models predicts for a person which activities are conducted when, where, for how long, for and with whom and the mode choices they will make to complete them. Some of these models rely primarily on econometric choice models based on the theory of the utility maximisation as described above while others use rule-based decision simulations.

There are in general two types of activity-based application model:

1. Tour-based models: initial type of activity-based model is the tour-based model. This approach uses a tour, a sequence of linked trips that begins and ends at the same location such as trip maker’s home, rather than a trip as the unit of analysis to capture the linkage between trips of the same tour. Although popular in the developed country cities practice, these kinds of models would be considerably more demanding in terms of data and model software requirements (Miller et al., 2005);

2. Activity scheduling models:

More recently, emphasis has shifted to activity scheduling and trip chaining behaviour of households. In the activity scheduling models, out-of-home activity schedules are explicitly generated to capture the processes of individuals implementing activity

(26)

20

decisions interactively with others, usually other household members, within a time constraint.

2.4 Land use and transport interaction models

Parallel to the development of the travel demand model mentioned above which leaves trip generation unaltered, urban planners also recognised the intricate interaction between the transport network and the rest of the urban system (Figure 2-1 from Southworth, 1995).

Figure 2-1 Complexity of functional linkages in urban system dynamics Source: Southworth, 1995

Changes in transport supply would influence land use configuration and in the long run, the re-configured land use would have a bearing on transport demand. Also, there are wider socio-economic changes which are not directly related to the transport supply and influence the configuration of land use. Therefore, an integrated framework, land use and transport interaction model (LUTI), has been developed in the 1950’s and 1960’s for evaluating complex urban systems in the current planning context. Unlike the conventional four-step model, which requires inputs of land use data forecasted exogenously, LUTI models forecast their own land use activities based on the input of land-use policies and the changes in accessibility brought about by variation in the transport system are reflected in the land use activities.

(27)

21

In terms of model structure, most operational LUTI models have three main sub-models. They are land use, socio-demographic and transport models. They are either fully integrated or loosely coupled with each model during the model execution (Figure 2-2).

Figure 2-2 Generalized structure of an operational LUTI model Source: Acheampong and Silva, 2015

The land use model usually includes information on the urban land market such as employment space ratio, land values, dwelling and occupancy types, housing vacancy etc. Most of the existing models have detailed urban land use and housing market sub-models. The socio-demographic model contains important socioeconomic variables that mediate household location choice and travel behaviour. Different model platforms capture varying levels of detail in terms of socio-demographic factors and processes. Most LUTI models divide households or population into segments of similar socioeconomic groups. For example, MUSSA–ESTRAUS (Martinez, 1996) and RAMBLAS (Veldhuisen, Timmermans, & Kapoen, 2000) are based on 13, and 24 different population segments respectively. Some models are able to capture the dynamics of the socio-demographic change. For example, DELTA-START (Simmonds, 1999) and UrbanSim (Waddell, 2002) has detailed demographic transition model that simulates the dynamics of household

(28)

22

formation, dissolutions, and transformations as well as employment transition model that simulates the creation and removal of jobs.

The transport model of most of the existing operational LUTI models, particularly the spatial interaction-based and utility-based ones, adopt the four-step approach as described earlier. As shown in Figure 2.2 above, the land use model is dynamically coupled with or, for some models, integrated within the transport model containing a network assignment module. The outputs from network assignment module are generalised transport costs, manifested by congested networks, travel times and distance, and are fed into the land use model.

In terms of modelling approaches,

Figure 2-3 describes an approximate timeline for adoption of various modelling approaches within transport and land use research.

(29)

23

Figure 2-3 Chronological development of LUTI models Source: Iacono, Levinson, & El-Geneidy, 2008

The transition from one approach to the other does not necessarily mean a complete abonnement of the previous approaches. Instead, new modelling approaches learn lessons from the past and adopt emerging theoretical and empirical insights aiming to address the limitations of their predecessors. Early LUTI models were aggregate spatial interaction models employing the gravity analogy supported by the theory of entropy maximisation (Wilson, 1970) to describe the behaviour of firms and households in space. Very few examples of this type of modelling framework remain. The shortcomings of these models were numerous. For example, most were static equilibrium models incapable of capturing the dynamics of urban systems; none of the models actually represented land markets with explicit prices; zones were highly aggregate and lacked spatial detail, and the models were inadequately supported by theory. Inadequate theory may have also been a reason that many of the models forecasted so poorly. Lee (1973) characterized mistakes of the first generation of models as being too complicated, overly aggregate, data hungry, wrongheaded, extraordinarily complicated, too mechanical, and expensive. Many of these criticisms informed the next generation of models, which took guidance from developments in econometric modelling based on random utility theory.

(30)

24

The work by McFadden (see Domencich and McFadden, 1975) and Williams (1977) gives the theoretical basis for modelling the spatial economy. As a result, economic evaluation is made possible through the use of random utility theory and advancements in discrete choice modelling.

LUTI models that follow econometric approach can be categorised as two types of models: regional economic models and land market models. In both models, transport flows are predicted by the core engine – a regional economic model realised by input-output analysis or a land market model of residential and commercial real estate. Several of the LUTI models with econometric approach continue to be used today since utility theory and theories of decision making under uncertainty are operationalized using mathematical formulations of mainly logistic regression models that vary in their complexity but are reasonably parsimonious and tractable (Iacono, Levinson, & El-Geneidy, 2008).

In the 1990s, with improvements in computational technology and advances in modelling theory and methodology, the microsimulation approach increasingly came into fashion, including activity-based models for travel, cell-based models for land use, as well as multi-agent models for urban simulation (Wegener, 2004). More recently, some researchers have begun to develop comprehensive urban microsimulation models that fully reflect the dynamics of changes in the population and the urban environment within which they make choices.

2.5 A case study of land use and transport model software MEPLAN

Of the main operational policy-oriented LUTI models, MEPLAN represents one of the most extensively developed, in terms of infrastructure planning assessments and land use/transport interaction mechanisms (Echenique, 2004). The development of MEPLAN first drew inspirations from the efforts to spatialize Leontief’s intersectoral Input-Output approach (Leontief, 1951, 1967). This approach provides a general framework to integrate intersectoral industrial activities, which was served as exogenous inputs of the urban dynamics in earlier Lowry-based models. Bringing the entropy theory and its logit forms, the approach extends into spatial disaggregation with origin-destination production and attraction-constrained coefficients. It has been used for planning researches in various cities (London, Maples, Helsinki, Tokyo, Cape Town and Santiago, Beijing), regions (the South East of England, the East Midlands region of England, the Basque region of Spain, Bolzano in Italy and the Central region of Chile), nations (Great

(31)

25

Britain, Colombia, Sweden, Chile and Argentina), and a strategic passenger and freight model of the European Union and surrounding countries.

2.5.1 Overall structure and operation of the model

The theoretical structure of the MEPLAN model is represented by three interrelated modules: the land use module (LUS), the land Use/transport Interface Module (FRED) and the transport module (TAS).

These modules operate on a time period by time period basis. This means the transport module is influenced by existing infrastructure and spatial patterns of activities; while the land-use module is affected by both the previous and current transport accessibilities. Figure 2-4 shows MEPLAN’s running through time.

Figure 2-4 Dynamic operation of the model

Figure 2-5 illustrates the typical structure and operation of a MEPLAN model and next section describes the role of the individual modules.

(32)

26

Figure 2-5 Typical structure and operation of a MEPLAN model

2.5.2 The land use module (LUS)

This module models the spatial location of activities such as employment and population and produces trades between zones. Planning information represented by groups is referred to as factors. These factors include elements such as population, employed persons, land/floorspace, goods vehicle travel, other travel and other goods and services. With the definitions of various factors, the land use submodel LUSA represents the spatial-economic linkages between activities or land uses.

(33)

27 LUSA incorporates the following:

(i) An input-output model. This is based upon the concept that the production of some economic activity (an output) consumes a range of other types of economic activities (inputs). These inputs, in turn, consume further inputs and so on during the process of production. To drive the input-output mechanism, MEPLAN relies upon the separation of employment into two types, an exogenous and endogenous employment. Exogenous employment is used to represent economic sectors producing primarily export demanded by markets outside the study area and the location of this employment is not determined by the need for access to a local market.

(ii) An elastic consumption model. This enables the consumption of goods, services and space to vary with prices and income. For example, households may consume more if incomes rise and businesses may use office space and labour more intensively as prices rise.

(iii) A spatial choice model. This predicts where factors will locate, and by extension the pattern of trades between zones. The spatial allocation process essentially takes as fixed the demand for a factor in a consumption zone, and distributes the satisfaction of this demand among the sources of supply in the production zones. The allocation process is based upon the "cost of living" in the zone, the disutility of travelling between the production and consumption zones, the availability of land/floorspace in the zone and an extra disutility (or attraction factor) term.

The land use sub-model is moved forward through time using a second sub-model, LUSB. This uses incremental models to allocate any zone specific or study wide land or changes in floorspace and exogenous employment. In the case of floorspace, LUSB allocates the increment to zones in proportion to zonal "attractiveness". This typically includes measures of zonal capacity for development, rent per unit land/floorspace for residential uses, and previous amount of land/floorspace and rental levels for retail and business uses. In the case of exogenous employment, allocation is also based on zonal attractiveness with the employment attractiveness term typically including previous employment and the cost of location (labour plus rents) in the zone.

2.5.3 The land use/transport interface module (FRED)

This converts land use trade matrices into transport flow matrices by socio-economic group and trip purpose or transport disutility into trade disutility.

(34)

28 2.5.4 The Transport module (TAS)

The main module of the transport sub-model, TASA, is comparable to the latter stages of a conventional four-stage transport model in that it carries out modal split, route assignment and capacity restraint. The trips are assigned into travel modes by logit models onto the transport networks. The assignment of traffic makes use of Dial’s probabilistic multipath assignment method (1971), which takes account of the costs and congested travel time over all forward feasible paths towards each origin. The resultant transport disutility are then passed on via the interface module to act as an influence on land use location in the next time period.

2.6 Case study 2: SIMULACRA

SIMULACRA is also prefaced by an input–output model which drives employment growth (Batty, et al., 2013). SIMULACRA is a series of fast, visually accessible, cross-sectional urban models for large metropolitan areas that enable the rapid testing of many different scenarios pertaining, both, short-term and long-term urban futures. The models are multi-sector, dealing with residential, services, and employment location. They are highly disaggregate, and subject to constraints on land use densities and transport capacities. Several versions of the model now exist, however, in here, a brief outline is presented. Figure 2-6 shows a basic outline of the cross-sectional structures and many-sector models that the framework dealt with.

This modelling effort began in 2007 with the construction of a residential location model for the Greater London Authority (GLA) region which was part of an integrated assessment of climate change, largely sea-level rise over the next 100 years. Later, the model extended to cover a much larger region including a region from Reading in the west to Southend in the east and from Luton in the north to Gatwick–Crawley in the south. This is what is referred to as the Outer Metropolitan Area which has a population of about 14 million in comparison with the GLA area that has around 8 million.

(35)

29 Figure 2-6 The SIMULACRA model structure.

Aggregate external totals are coloured dark grey; external zonal variables are coloured light grey; predicted variables are coloured white.

Source: Batty et al., 2013

The model links activity types through spatial interactions: the journey home to work defined by trips 𝑇𝑖𝑗 linking population to employment, trips 𝑆𝑖𝑗 from residential areas to services centers (shopping, health, education, and leisure), and through implicit industrial linkages measured as accessibilities to employment and to commercial activities. A formal description of these activities and the way they can be disaggregated and extensions to such classifications are shown in Figure 2-7 and Table 2-1.

(36)

30

Figure 2-7 SIMULACRA framework overall sequence of sub-models Source: Batty et al., 2013

(37)

31 2.7 Critical summary

Conventional four step model do not generally tend to model trips outside AM peak period, or even if they did, the results are rarely well checked. Activity-based travel demand models explicitly treat spatial and temporal interdependencies in activity and travel choices and therefore discretionary travels occurring outside the AM peak period are inherently modelled. Also, this type of model views travel within the context of overall daily time-use and emphasis on household level decision making and interaction among household members. So, it is suitable for modelling the time-of-day choices which help assess travel demand management strategies such as congestion charging. However, such models although lends itself well to realistic policy evaluations, they do not consider the wider economic and land use factors that may have an overarching effect on the congestion level. The impacts of wider economic and land use factors are especially an urgent matter for urban hubs where the business, social and leisure activities aggregate and crowding and congestion occurs outside the AM peak period. As the overcrowding and congestion translate into high property and goods prices, lower productivity, inefficiency and social exclusion may ensue.

As reviewed earlier, LUTI model provides a framework to model the interaction between the land use and transport that allows for forecasting medium to long term travel demand changes. It can potentially address the deficiency of the traditional four step model and activity-based model that the land uses are seen as fixed. While the early LUTI models were not in use due to several reasons described above the models based on microsimulation have too high data requirement and long running time and therefore has rarely been used in practice. Moreover, the stochastic variation in model results between different simulations runs of different random number seeds means that model results may be presented with illusionary precision (Wegener, 2011). Several of the LUTI models with econometric approach such as MEPLAN continue to be used today grounds in robust economic and behaviour theories while the mathematical formulations are tractable. However, such types of LUTI models tend to focus on non-discretionary activities such as rush-hour commuting. The existing models are not well adapted to addressing non-discretionary social and leisure activities such as during afternoon and evening periods. In MEPLAN, the personal trips for purposes other than commuting and employer’s business are currently modelled in a relatively simple way. Also, no differentiation has been made regarding different leisure activities, in terms of trip attractions. Similarly,

(38)

32

SIMULACRA does not directly model shopping and leisure trips in its spatial input-output structure. Instead, it applies a conversion factor to derive shopping trips based on the estimated retail jobs at a rather crude aggregate level. The model although applies in Greater London Area, it does not model night time activities in London where the night economy has become an essential part of the city. Notwithstanding the above, the model’s design to allow quick and agile scenario tests is unique in the model of this kind.

To sum up, the principle of activity based modelling provides a way to generate discretionary travel demand such as night market visits. MEPLAN has been developed based on random utility theory which reflects individual’s travel behaviour and the econometric framework provides the opportunity to estimate travel demand from potentially a wide range of land use activities, which is increasingly supported by emerging social media sources. The simple and direct SIMULACRA style of scenario tests with simple, incremental input changes enables many different alternative futures to be assessed rapidly, particularly for strategic assessments of distinct land use and transport scenarios.

(39)

33

Chapter 3 METHODOLOGY

3.1 Methodological framework

The purpose of this research is to establish a new methodological framework for modelling traffic in all time periods in a large city region for practical policy applications. This means that the method will draw upon all three main transport modelling traditions and then extend them in a way that can be well supported by currently available data sources. The methodological framework will thus combine three intellectual traditions of transport demand and supply modelling: first, it will adopt the mode choice and traffic assignment modules from the four step transport model tradition because this is what can be supported by current data sources in model calibration and validation; second, for generating discretionary travel demand such as residents’ and tourists’ night market visits, it will adopt the principles of activity based modelling, where we establish new methods that are supported by emerging social media sources; thirdly, for forecasting medium to long term travel demand changes, it adopts a land use/transport interaction modelling framework; more specifically it sets up the core travel demand forecasting capabilities in the MEPLAN tradition, and adopts the quick and agile SIMULACRA style of scenario tests with simple, incremental input changes. The modelling framework can thus become capable of generating medium to long term travel demand shifts and translate them into modal flows and traffic assignments across all the peak and non-peak time periods. Such model outputs can then be used through pairwise scenario comparisons of distinct land use and transport policy packages and provide insights into the efficacy of infrastructure investment as well as the effectiveness of travel demand management. Figure 3-1 illustrates the core model of the methodological framework, which builds on the land use/transport interaction models. This core model is established for each modelled year whether in the base year (for calibration, verification and validation) or the forecast years.

(40)

34 Figure 3-1 Proposed model structure

The information flow within the core model can be broadly described in terms of four interlinked simulation procedures, the first two of which are considered land use and the remaining two transport:

(a) Activity generation: Households of different socio-economic groups in the consumption zones are used to generate demand for education, shopping/personal business, leisure, and other services, which are provided in the production zones. Similarly, demand for night market activities, located in the production zones such as the night market clusters, are estimated based on the presence of residents and tourists. (b) Spatial distribution of the travel demand: Logit-based discrete choice models are applied to distribute the demand between each pair of production and consumption zones.

(41)

35

(c) Mode Split: The journeys are attributed to modes of transport that are available between each pair of origin and destination zones, according to the modal choice behaviour of each socio-economic group, for each travel purpose, for each time period. (d) Link Assignment: The journeys on each mode are then assigned to road and rail (including metro) networks using a stochastic user equilibrium algorithm, which can either incorporate road and rail service capacity restraints, be assigned free of capacity restraints or be assigned based on certain assumptions of traffic speeds (such as maintaining the constancy of congestion as observed in the base year) in order to explore future utilisation of the transport network and services under alternative planning scenarios.

Whilst travel demand information is generated and fed top-down from (a) to (d), the travel costs, times and time based generalised costs are transmitted bottom-up from (d) to (a). This allows an extensive interaction between land use and transport. In the above, Steps (a) and (b) are regarded as the land use module, which replaces the trip generation and trip distribution steps of a conventional four-step model. Steps (c) and (d) are thus the main contents of the strategic transport module.

3.2 Land use

For the purposes of the model, activities will be split into two broad groups:

(a) conventional activities that will translate into trips including Home Based Work (HBW), Home Based Education (HBE), Home Based Other (HBO) and Non Home-Based (NHB)

(b) night market activities that will translate into trips to night market clusters, which are rarely covered by existing transport or land use/transport models.

Within the model, each sector is composed of a number of land use factors which represent various activities of production and consumption and respective producers and consumers.

(42)

36 3.2.1 Conventional activities

3.2.1.1 Conventional activity generation

The demand for conventional activities is generated based on households, which represent the consumers in the model. The demand can be expressed as follows:

𝑌𝑗𝑚𝑛 = 𝐻𝐻𝑗𝑚∗ 𝑡𝑚𝑛 [Equation 3.1] Where

𝑌𝑗𝑚𝑛 is the total demand for factor n by household type m in zone j

𝐻𝐻𝑗𝑚 represents the number of household type m in zone j

𝑡𝑚𝑛 represents the demand coefficient for factor n by household type m

3.2.1.2 Spatial interaction process of conventional activities

The households’ demands for industrial and service factors are transportable (trade) and the model estimates the location of production of these factors. The trade from zone i to zone j is equal to the total demand for a factor in zone j multiplied by the probability of it being produced in zone i. A logit-based discrete choice model is applied to simulate the spatial distribution for these trades:

𝑇𝑖𝑗𝑚𝑛 = 𝑌𝑗𝑚𝑛 𝑆𝑖𝑛𝑒−𝜆𝑚𝑛(𝑑𝑖𝑗

𝑚𝑛)

∑ 𝑆𝑖 𝑖𝑛𝑒−𝜆𝑚𝑛(𝑑𝑖𝑗𝑚𝑛)

[Equation 3.2]

where

𝑇𝑖𝑗𝑚𝑛 is the flow of factor n (in production zone i) demanded by household type m in consumption zone j

𝑌𝑗𝑚𝑛 is the total demand for factor n by household type m in zone j

𝑆𝑖𝑛 is a factor influencing the attractiveness of factor n in zone i

𝑑𝑖𝑗𝑚𝑛 is the disutility of transport (generalised cost) between zone i and j for factor n demanded by household type m

𝜆𝑚n is a positive concentration parameter which specifies the choice behavior of factor n

(43)

37 Attractiveness of conventional activities

The term influencing the attractiveness of factor can be expressed as follows (CECI, 2009):

𝑆𝑖𝑛 = 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡𝑛 + 𝑎𝑛𝐸1

𝑖 + 𝑏𝑛𝐸2𝑖+ 𝐶𝑛𝑃𝑖 [Equation 3.3]

Where

𝑆𝑖𝑛 is a factor influencing the attractiveness of factor n in zone i 𝐸1𝑖 is number of employees in a secondary industry in zone i 𝐸2𝑖 is number of employees in a tertiary industry in zone i 𝑃𝑖 is the number of students in zone i

𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡𝑛, 𝑎𝑛 , 𝑏𝑛and 𝑐𝑛 is the parameters for different types of factor n

3.2.2 Night market activities

Since the night markets have recently become a major tourist attraction in cities around the world, it is worth factoring in the tourists when it comes to simulating activities occurring at night. Therefore, the night market activities consist of those of residents and those of tourists.

3.2.2.1 Night market activity generation by residents

According to the literature (Yen, 2011; Lee et al., 2015), it is found that night markets visitors differ greatly by age group. Therefore, the demand for night market activities is generated based on people of different age groups in the model.

𝑌𝑗𝑎 = 𝑃𝑜𝑝𝑗𝑎∗ 𝑡𝑎 [Equation 3.4]

Where

𝑌𝑗𝑎 is the total demand for services of the night market activities by residents of age group a in zone j

𝑃𝑜𝑝𝑗𝑎 represents the number of population of age group a in zone j

Figure

Updating...

References

Updating...